Test and monitor
Test an agent in the browser
Check text, microphone audio, interruptions and action results before deployment.
Text first, then audio
- Save the agent draft.
- Select Test in the editor.
- Start a text conversation to check instructions and facts. Text tests do not execute connected tools.
- Start an audio test and allow microphone access.
- Speak a complete scenario, including a correction and an interruption.
- End the session and inspect its transcript and outcomes.
Use a current browser with microphone permission and an HTTPS page. Close another app that is using the microphone if capture fails. A blocked autoplay policy may require clicking the call or play control before sound begins.
Suggested receptionist scenarios
| Scenario | Expected result |
|---|---|
| Ask for Saturday hours | Agent uses approved clinic information. |
| Give and correct a pet name | Agent uses the correction later. |
| Request an unavailable time | Agent checks alternatives rather than inventing a slot. |
| Ask to cancel | Agent verifies the booking and waits for a successful receipt. |
| Interrupt a long answer | Playback stops and the agent responds to the new request. |
| Ask for staff | Agent follows the configured handoff policy. |
Audio tests leave tool execution off by default. To validate real booking or HTTP actions, open Test settings, enable Execute connected tools during this test, and start a fresh audio session. Those tools can change connected systems. Use a sandbox or isolated test calendar, identify test records clearly and remove them in the source system afterward. A text test or an audio test with tools disabled cannot prove a real booking was created.
A web call does not establish a verified phone caller number. It cannot prove PSTN audio, carrier caller ID, DTMF, SMS delivery or human phone-transfer behavior. Complete those tests on the authorized deployment before changing the clinic's public routing.